摘要
股票时间序列预测在经济和管理领域具有重要的应用前景,也是很多商业和金融机构成功的基础.首先利用奇异谱分析对股市时间序列重构,降低噪声并提取趋势序列.再利用C-C算法确定股市时间序列的嵌入维数和延迟阶数,对股市时间序列进行相空间重构,生成神经网络的学习矩阵.进一步利用Boosting技术和不同的神经网络模型,生成神经网络集成个体.最后采用带有惩罚项的半参数回归模型进行集成,并利用遗传算法选择最优的光滑参数,以此建立遗传算法和半参数回归的神经网络集成股市预测模型.通过上证指数开盘价进行实例分析,与传统的时间序列分析和其他集成方法对比,发现该方法能获得更准确的预测结果.计算结果表明该方法能充分反映股票价格时间序列趋势,为金融时间序列预测提供一个有效方法.
A novel neural network ensemble model is proposed for stock market prediction. First of all, the original data of time series are reconstructed for reduction the noise and extraction the tendency by Singular Spectrum Analysis （SSA）. Secondly, C-C algorithm are adopted to confirm the best delay time and the best embedding dimension for phase space reconstruction, and the learning matrix can be obtained. Third, many individual neural networks are generated by Bagging techniques and different model of neural network. Finally, the nonparametric regression （NR） model is used to neural network ensemble based on Gaussian kernel function estimation with variable band-with. This method be established the forecast model of Stock Market by the opening price of S＆P 500 index as an example, more accurate results can be acquired by this method compared with the traditional time series analysis and other integrated methods. The result shows the way have high accuracy, and it is a useful tool for the stock market forecasting.
出处
《数学的实践与认识》
CSCD
北大核心
2012年第11期58-68,共11页
Mathematics in Practice and Theory
关键词
奇异普分析
相空间重构
神经网络集成
非参数回归
singular spectrum analysis
phase space reconstruction
neural network ensemble
semi-nonparametric regression.
